Privacy of outsourced k-means clustering
Dongxi Liu, Elisa Bertino, Xun Yi · 2014
It is attractive for an organization to outsource its data analytics to a service provider who has powerful platforms and advanced analytics skills. However, the organization (data owner) may have concerns about the privacy of its data. In this paper, we present a method that allows the data owner to encrypt its data with a homomorphic encryption scheme and the service provider to perform k-means clustering directly over the encrypted data. However, since the ciphertexts resulting from homomorphic encryption do not preserve the order of distances between data objects and cluster centers, we propose an approach that enables the service provider to compare encrypted distances with the trapdoor information provided by the data owner. The efficiency of our method is validated by extensive experimental evaluation.